Embed Size (px)
Citation preview
Journal of Memory and Language, In Press
Fast Morphological Effects in First and Second Language Word Recognition
Kevin Diependaele (1)
Jon Andoni Duñabeitia (2)
Joanna Morris (3)
Emmanuel Keuleers (1)
1. Ghent University; Ghent, Belgium.
2. Basque Center on Cognition, Brain and Language (BCBL); Donostia, Spain.
3. Hampshire College; Massachusetts, U.S.A.
Kevin Diependaele
Department of Experimental Psychology
Henri Dunantlaan 2
9000 Ghent
Belgium
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% 6%
Abstract
In three experiments we compared the performance of native English speakers to that of
Spanish-English and Dutch-English bilinguals on a masked morphological priming lexical
decision task. The results do not show significant differences across the three experiments. In line
with recent meta-analyses, we observed a graded pattern of facilitation across stem priming with
transparent suffixed primes (e.g., viewer-view), opaque suffixed or pseudo-suffixed primes (e.g.,
corner-corn) and form control primes (e.g., freeze-free). Priming was largest in the transparent
condition, smallest in the form condition and intermediate in the opaque condition. Our data
confirm the hypothesis that bilinguals largely adopt the same processing strategies as native
speakers (e.g., Lemhöfer et al., 2008), and constrain the hypothesis that bilinguals rely more
heavily on whole-word processing in their second language (Ullman, 2004, 2005; Clahsen et al.,
2010). The observed pattern of morphological priming is in line with earlier monolingual studies,
further highlighting the reality of semantic transparency effects in the initial stages of word
recognition.
Keywords: Morphological processing, bilingual word recognition, masked priming, semantic
transparency.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% 7%
Introduction
Over the last decade we have witnessed an exponential growth in studies investigating the
role of morphology in visual word recognition using the masked priming paradigm (Forster &
Davis, 1984). When target words are preceded by a morphologically related prime (e.g., worker-
work), target recognition (typically measured using the lexical decision task) occurs faster relative
to form controls (e.g., freeze-free) and semantic controls (e.g., giraffe-safari). In masked priming,
participants are unaware of the primes, which are presented (a) immediately before targets, (b)
between masks (e.g., hash signs) and (c) for around 40-50ms (see Diependaele, Grainger &
Sandra, 2010, for a review). The results of masked morphological priming experiments suggest
that morpheme-sized representations (shared by primes and targets) are automatically activated at
a very early stage of visual word recognition. Interestingly, this does not exclusively depend on
whether or not primes and targets share their stem (e.g., worker-work). Recent evidence has
shown that morphological information is also taken into account when masked primes and targets
share an affix (e.g., darkness-happiness vs. shallow-follow; Duñabeitia, Perea, & Carreiras, 2008;
Chateau, Knudsen, & Jared, 2002), highlighting the visual word recognition system’s reliance on
morphological information.
A key result in stem priming is that morphological facilitation also occurs when targets are
preceded by semantically opaque and pseudo-complex affixed masked primes (e.g., department-
depart and corner-corn; see Rastle & Davis, 2008, for review)(footnote 1). This finding has
revived so-called “blind decomposition” or sublexical accounts of morphological processing. In
these accounts morphological activation occurs at the level of sublexical form representations.
Inputs are parsed into morphemes, without any reference to whole-word lexical information (e.g.,
Taft & Forster, 1975). Further evidence for this view comes from the observation that newly
constructed (i.e., unfamiliar) suffixed forms (e.g., cornly) elicit facilitation similar to that of
familiar transparent primes, regardless of their interpretability (Longtin & Meunier, 2005, but see
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% 8%
Morris, Porter, Grainger & Holcomb, 2010). Morphological activation thus occurs regardless of
semantic transparency and whole-word familiarity, two clearly lexical variables.
It remains the case, however, that priming effects are somewhat larger for semantically
transparent familiar derivations. Feldman, O’Connor and Moscoso del Prado Martin (2009)
showed that when the data of individual studies are combined in a meta-analysis, semantically
transparent primes show a significant advantage over opaque or pseudo-complex primes, whereas
this advantage is not always significant in the individual analyses (see also Davis & Rastle, 2010).
In addition to their meta-analysis, Feldman et al. (2009) also conducted a new experiment in
which they observed a clear semantic transparency effect (i.e., priming advantage of transparent
over opaque items) with carefully controlled stimuli, in line with earlier results of Diependaele,
Sandra and Grainger (2005, 2009) and Morris, Franck, Grainger and Holcomb (2007).
One way of explaining graded priming across transparent and opaque items is to assume
that the morphological representations that are activated during sublexical processing, i.e., the
morpho-orthographic (and/or –phonological) representations, rapidly send activation to
corresponding whole-word representations, viz., the lexical representation of the full input and the
stem (e.g., worker, work; corner, corn). These lexical representations could be form
representations (e.g., Diependaele et al., 2005, 2009), more abstract lemma representations (e.g.,
Taft & Nguyen-Hoan, 2010) and/or whole-word semantic representations. The critical point is
that if the competition among lexical these representations is somehow mediated by the (morpho-
)semantic relationship, such that competition is smaller for transparent derivations than for
opaque or pseudo-derivations, the activation of stem representations could rapidly become larger
for semantically transparent familiar derivations. Giraudo and Grainger (2000) and Diependaele
et al. (2005, 2009), for instance, proposed that the competition between lexical form
representations is mediated by positive feedback from higher-level morpho-semantic (so-called
“supra-lexical”) representations (footnote 2). Taft and Nguyen-Hoan (2010) more recently
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% 9%
proposed that sublexical morpho-orthographic representations map onto lemma representations.
These representations are connected through positive links in the case of semantically transparent
morphological relatives. No such links are present between the lemma representation of opaque
derivations and their (pseudo-)relatives.
The graded pattern across transparent and opaque items constitutes simultaneous evidence
for sublexical morphological processing (i.e., priming from opaque or unfamiliar derivations) and
lexical morphological processing (i.e., larger effects with semantically transparent familiar
derivations). Hence, there appears to be no immediate elimination of the sublexical morphemic
activation in the case of opaque or unfamiliar complex words once (morpho-)semantic properties
come into play. This could indicate a parallel architecture where lexical morphological processing
is initiated through direct whole-word activation, while bypassing - and not immediately
interfering with - sublexical morphological segmentation (e.g., Diependaele et al., 2005). It is also
possible to account for this in a sequential view, however. Taft and Nguyen-Hoan (2010) consider
for instance (among other possibilities) that: “… although the lemma for corn and corner vie with
each other to reach a recognition threshold, such competition does not actually inhibit the
activation level of the ‘loser’. That is, all that matters is which lemma reaches threshold first.” (p.
291). The simultaneous evidence for sublexical and lexical morphological processing thus does
not necessitate a parallel architecture. It can be argued that the sequential view nevertheless
implies an initial stage in processing, where morphemic activation is purely morpho-orthographic
(i.e., sublexical) and morphological priming should therefore be matched across different levels
of semantic transparency and familiarity. This prediction contrasts with the data of Diependaele et
al. (2005, Experiment 2) who observed earlier stem priming with semantically transparent than
with semantically opaque suffixed primes: in an experiment with different prime durations.
Again, this pattern is in line with a parallel architecture, but it does not necessitate it. In the
sequential view, assuming cascaded processing allows that morpho-semantically mediated lexical
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% ;%
competition already starts before sublexically activated morphemic representations have reached
a criterion activation level required for observing priming with brief prime exposure.
In the present study we consider the processing of morphologically complex words from a
different angle, i.e., second language (L2) processing. In a recent review, Clahsen, Felser,
Neubauer, Sato and Silva (2010) claimed that current evidence shows that adult L2 learners are
not as sensitive to morphological information as native speakers, a statement based primarily on
their own research and on studies with inflected forms. In a study by Silva and Clahsen (2008) no
masked stem priming effects were found for L2 learners of English with regular past-tense primes
(e.g., boiled-boil), whereas these effects were observed for native English speaking participants.
Neubauer and Clahsen (2009) tested native and non-native German speakers in a series of
experiments exploring the processing of irregular and regular participles. Critically, in
Experiment 3 they used the masked priming procedure and showed that while facilitative priming
effects were found for irregular and regular participles in the L1 group, non-natives exclusively
showed priming effects for the irregular participles, concluding that L2 learners do not segment
inflectional affixes from their stems during processing. The explanation that is provided by these
authors is that, in general, the later in life words are acquired, the more their processing will rely
on direct lexical retrieval instead of grammatical computation. This idea derives from the
declarative/procedural model proposed by Ullman (2004, 2005). In this model, word recognition
depends on two distinct memory subsystems: declarative memory, situated in the temporal lobe,
and procedural memory, located in the frontal cortex and basal ganglia. The declarative system
provides mechanisms to store and access whole-word representations. The procedural system on
the other hand, provides mechanisms to acquire and use grammatical rules. Through this system,
regular inflections do not have to be stored in separate whole-word representations, as they can be
recognized and produced by applying combinatorial rules. However, through maturation, the
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% <%
declarative system becomes dominant later in life. As such, pure combinatorial forms in a non-
native language will nevertheless acquire a whole-word representation. They are not handled by
the procedural system, hence eliminating morphological priming in non-native languages.
Support for this explanation remains scarce. Neubauer and Clahsen (2009) did observe
priming for non-native irregular inflections. It is difficult to see how the proposed maturation
would differentially affect the processing of regular and irregular inflections. The implications of
this view for derived words also remain unclear. The study of Silva and Clahsen (2008) did not
show a similar elimination of stem priming in L2 with suffix-derived primes (boldness-bold) as
was observed with past-participle primes (boiled-boil). There was nevertheless a noticeable
reduction compared to the effects in L1. The authors concluded that “L2 learners employ
morphologically structured representations for derived word forms during processing, albeit less
effectively than native speakers” (p. 257). Hence, there appears to be some evidence for a general
reduction in the use of morphology in L2 word recognition. It is difficult to explain that
morphological effects are absent with inflection, but somewhat present with derivation, if one
assumes a common origin for both (e.g., Bybee, 1985; Raveh & Rueckl, 2000). Results in L1
commonly show stronger morphemic activations with inflections than derivations (e.g., Feldman,
1994; Schriefers, Frederici & Graetz, 1992, see also Silva & Clahsen, 2008). Hence, one would
expect the effects to be eliminated first in the case of derivations. The data seem to reveal the
opposite pattern. Clahsen, Sonnenstuhl and Blevins (2003; see also Silva & Clahsen, 2008) have
proposed an account that “treats productive inflection and derivation both as the result of
combinatorial operations but associates productive derivation (like irregularly inflected items)
with stored entries” (p. 24). In this approach, derivations are stored as whole-words with
reference to their internal morphological structure, whereas inflections do not have a whole-word
representation. Inflections are produced and recognized purely on the basis of stored rules. Even
within this model, where morphological effects for inflections and derivations arise from
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% =%
qualitatively different mechanisms, it is not straightforward to explain the current data pattern.
Following the declarative/procedural model, only the use of the rule-based route should be
reduced in L2, which predicts no reduction of morphological effects in the case of derivations.
As Clahsen et al. (2010) also note, current data about L2 morphological processing is very
limited. Silva and Clahsen (2008) only considered suffix-derived primes with the suffixes –ness
and –ity, considerably limiting the scope of these findings. A recent study by Feldman, Kosti!,
Basnight-Brown, Filipovi!-"ur#evi! and Pastizzo (2009) also sheds a different light on the
findings with inflections. In line with Silva and Clahsen (2008) and Neubauer and Clahsen
(2009), late bilinguals only showed facilitated stem recognition in L2 with irregular past participle
masked primes (Experiment 1; taught-teach). However, only the least proficient bilinguals
showed this pattern and it was not found for participles that preserve stem letter length (fell-fall).
Across different proficiency levels, bilinguals displayed highly similar effects as compared to
native participants, at least when facilitation was assessed relative to an unrelated baseline (billed-
bill vs. careful-bill). Relative to an orthographic control condition (billion-bill), the lowest
proficient bilinguals showed no evidence for morphological effects, whereas the highest
proficient participants only showed significant facilitation with regular past participle primes
(billed-bill). Feldman et al. suggested that at relatively low L2 proficiency levels, there is a
greater reliance on word forms and an impaired access to semantics. In some way, the
undistinguishable masked orthographic and morphological effects could be interpreted by the
declarative/procedural model, as showing an over-reliance on the declarative memory system, and
poor compositional strategies, or under-reliance on the procedural memory system. However, the
results for highly proficient bilinguals are at odds with to those obtained by Neubauer and
Clahsen (2009).
It is thus far from clear what differences really do and do not exist between L1 and L2
morphological processing, and how these potential disparities might depend on the linguistic
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% >%
proficiency level of non-native speakers of a language. In this context, the present study sets out
to provide a detailed test of the differences in the processing of suffix-derivations in L1 and L2.
We adopt a large-scale masked morphological priming design in English (25 items per condition)
with a group of native English speaking participants (n=65) and two groups of relatively
proficient L2 participants: Spanish-English bilinguals (n=66) and Dutch-English bilinguals
(n=65). As can be seen in Table 3, the two groups rated their L2 proficiency differently with
respect to age of acquisition, age of relative proficiency, comprehension skill, reading skill and
exposure. We can thus assess how well our conclusions generalize across bilinguals with different
language backgrounds. Like in previous studies, we consider stem priming in the lexical decision
task with semantically transparent and opaque derivational suffixed primes relative to a condition
with form control items (cf. Rastle & Davis, 2008). We do not limit ourselves to any particular set
of suffixes as opposed to those used in the study of Silva and Clahsen (2008). The comparison
between semantically transparent and opaque items allows us to assess whether any L1-L2
differences interact with semantic transparency. Finally, the comparison with pure form priming
allows us to test whether any reduction in the size morphological priming is accompanied by a
stronger reliance on orthographic representations or on word forms in a non-native language,
which would support the hypothesis of Feldman et al. (2009).
Experiment 1
Experiment 1 serves as the baseline for our study of L2 morphological processing. Native
English speakers were tested with the English materials. We expected to find the general pattern
found in the literature (cf. Rastle & Davis, 2008; Feldman et al., 2009): greater priming for
transparent and opaque items relative to form controls and a (potentially weak) advantage of
transparent over opaque primes.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% 4?%
Method
Participants
Sixty-five native speakers of English participated in Experiment 1. They were recruited
from the Hampshire College community in the USA. Their age ranged from 18 to 32, with a
mean of 22. There were 39 females and 26 males in the sample. Each participant reported having
normal or corrected-to-normal vision.
Materials and Design
Words. 150 English words served as word targets in the experiment. Each target was
paired with two primes: a related and an unrelated word. The pairs were selected from the large
collection of similar experiments now available in the literature. Unrelated primes were unrelated
to the target in both form and meaning. They were always suffixed words matched to the related
primes on length and frequency. Related primes were either a transparent suffixed morphological
relative of the target (e.g., viewer-view; n=50), a word that could be parsed into the target plus a
suffix (e.g., corner-corn; n=50) or a word that could be parsed into the target plus a non-
morphemic word ending (e.g., freeze-free; n=50). Relevant distributional characteristics of the
related primes and targets are available in Table 1. There was always a maximal initial letter
overlap between related primes and targets, i.e., a target was always fully embedded at the
beginning of its related prime. We tested the manipulation of semantic transparency with Latent
Semantic Analysis (LSA; Landauer & Dumais, 1997; http://lsa.colorado.edu). There was no
significant difference between the LSA score for related prime-target pairs in the form and the
opaque condition (.10 versus .09 respectively; Welch t < 1), whereas there was a significantly
larger score in the transparent condition (.39; Welch t (62.80) = 7.69, p < .001 compared to the
opaque condition and Welch t (62.28) = 7.60, p < .001 compared to the form condition). All word
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36% 44%
stimuli are available at http://users.ugent.be/~kdiepend/supp/bilingmorphstim.txt. We created two
balanced experimental lists by rotating Relatedness within the levels of Prime Type using a Latin
square design. These lists were distributed evenly across the participants.
!
!
"#!Table 1. Length and frequency characteristics of the related primes and targets. Frequencies (log10 scale) were obtained from the CELEX English
lexical database (Baayen, Piepenbrock & Gulikers, 1995). Token frequencies were computed per million. Welch t is provided for significant
condition differences.
Transparent Opaque Form Transparent-Opaque Transparent-Form Opaque-Form
Primes Number of characters 6.56 6.50 6.44
Surface frequency 0.74 0.82 0.73
Suffix family frequency 3.45 3.22 n.a. t(83.66) = 2.12, p < .05
Suffix family size 2.87 2.63 n.a. t(89.27) = 2.22, p < .05
Suffix form frequency 4.09 4.01 3.02 t(70.44) = 5.92, p < .001 t(73.11) = 5.41, p < .001
Boundary frequency 4.08 4.14 3.91
Neighborhood size 6.62 7.80 9.78
Neighborhood frequency -0.28 0.03 0.20
Targets Number of characters 4.30 4.16 4.09
Surface frequency 1.46 1.57 1.42
Family frequency 1.98 1.93 1.76
Family size 1.18 1.11 0.95 t(94.23) = 2.79, p < 0.05
Neighborhood size 31.96 33.60 43.16
Neighborhood frequency 2.01 2.24 2.04
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
47%
Nonwords. Each word list was coupled with the same nonword list, consisting of 150
nonword targets. The targets were constructed by altering one or two letters of existing English
words, while making sure that the resulting strings were regularly pronounceable in English and
that their length was similar to that of the word targets. Half of the nonword targets were paired
with an orthographically related suffixed word (purely-gure), while the other half were paired
with an unrelated suffixed word (penalty-murf). These primes were drawn from the same
frequency range as the word target primes and also had a similar length.
Procedure
The experiment was controlled by the DMDX software package (Forster & Forster, 2003).
All visual stimuli were presented in 12pt fixed width font (Courier New). Participants were
instructed to focus on the middle of the screen at the beginning of each trial. A trial began with
the central presentation of a forward mask (a row of hash signs). After 500ms the mask
disappeared and the prime was presented for 53ms (4 refresh cycles of a 75Hz CRT monitor). The
prime was immediately followed by the target in upper case, which stayed on the screen until the
participant pressed one of the two response buttons or after a deadline of 2500ms. Reaction times
were recorded using a two-button response box. Participants were asked to decide as quickly and
accurately as possible whether or not the target corresponded to an existing English word by
pressing the corresponding response button. The choice of buttons for word and nonword
responses was left up to preference. Targets were presented in a different random order for each
participant. The experiment started with four practice trials.
Results and Discussion
We analyzed the correct RTs and accuracies (95%) for word targets with linear mixed-
effects (lme) models with participants and items as crossed random variables (cf. Baayen,
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
48%
Davidson & Bates, 2008) as implemented in the lme4 package (Bates & Maechler, 2009) in R (R
Development Core Team, 2009). For accuracies, we used a generalized lme with logistic link
function. There was no averaging of the data prior to the analyses. We inverse-transformed all
RTs (i.e., -1000/RT) to reduce the positive skew in the distributions. Transformed RTs smaller
than Q1-3∗IQR or larger than Q3+3∗IQR, by either participants or items (0.1%), were excluded
from the analyses (with Q1 the first quartile, Q3 the third quartile, and IQR the interquartile
range). Missing RTs were replaced on a by-participant basis using Multiple Iterative Regression
Imputation (as implemented in the mi package; Gelman, Hill, Yajima, Su & Pittau, 2009)
(footnote 3). Condition means are presented in Table 2. Significance values were obtained
through the Markov Chain Monte Carlo (MCMC) sampling method (sample size = 10,000) for
the RT data. In each analysis, we first look at the interaction of Relatedness and Prime Type in the
ANOVA (footnote 4). If significant, we evaluate individual priming effects by checking the
model’s estimates for the contrast-coded levels of our design (footnote 5). If not significant, we
eliminated Prime Type from the model to test the overall effect of Relatedness. Model estimates,
along with p-values and MCMC-based confidence intervals are available at
http://users.ugent.be/~kdiepend/supp/bilingmorphlmer.pdf.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
49%
Table 2. Mean reaction times and accuracy proportions per condition in Experiment 1.
Prime Type Relatedness RT Accuracy
Form related 636 0.94
unrelated 637 0.93
effect 1 -0.01
Opaque related 612 0.95
unrelated 627 0.94
effect 15 -0.01
Transparent related 592 0.97
unrelated 628 0.95
effect 36 -0.02
Relatedness interacted significantly with Prime Type (F(2, 9744) = 8.06, p < .001) in the
RT data. As can be seen in Table 3, priming effects were largest in the transparent condition
(36ms; t = 7.51, p < .001) and smallest in the form condition (1ms; t = 1.93, p = .07; interaction: t
= 3.94, p < .001). Opaque primes elicited a priming effect in between that of the transparent and
form items (15ms; t = 3.81, p < .001). However, only the difference with the transparent condition
reached significance (t = 2.62, p < .05 versus t = 1.32, p = .18).
The accuracy analysis showed no significant interaction of Relatedness and Prime Type.
There was nevertheless a significant overall effect of Relatedness (z(9747) = 2.43, p < .02; 1%
more accuracy following related primes).
The pattern in Experiment 1 converges with earlier masked morphological priming results
in L1 (cf., Rastle & Davis, 2008; Feldman et al., 2009; Davis & Rastle, 2010). We find significant
priming for both transparent and opaque items, but not for form controls. The effect in the opaque
condition is located in the middle of the form and transparent condition effects (cf. Table 2). The
presence of significantly larger facilitation for transparent items is in line with the meta-analyses
in Feldman et al. (2009) and Davis and Rastle (2010). Our statistical tests nevertheless diverge
from earlier studies regarding the advantage of opaque items over form items, as it fails
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
4:%
significance (p = .18). Strictly speaking, this means that we do not find sufficient (i.e. p < .05)
evidence for a morphological effect with opaque items. It is important to consider normal
between-experiment variability here, however. Inspecting the literature overview of Rastle and
Davis (2008), the opaque-form difference in Experiment 1 (14ms) is clearly not an outlier
(median = 16, Q1 = 12, Q3 = 18, mean = 20, sd = 20). The distance between the present difference
and the sample median and mean corresponds to approximately one third of the IQR (6ms) and
the sd, respectively. The transparent-opaque difference (21ms) is relatively large, but the
distribution of this difference (median = 4, Q1 = -2, Q3 = 13, mean = 6, sd = 11) indicates that it is
safest to conclude that the pattern in Experiment 1 fits the graded pattern in the recent meta-
analyses (e.g., median+2.5*IQR = 42ms; mean+2.5*sd = 32ms), showing significant
morphological effects with both transparent and opaque items together with a significant
advantage for transparent items. We thus argue that Experiment 1 provides an adequate baseline
for comparing masked morphological priming across L1 and L2. The first comparison is with
Spanish participants with English as a second language. We return to the issue of between-
experiment variability when performing a joint analysis of our Experiment 1, 2 and 3.
Experiment 2
Method
Participants
Sixty-six native speakers of Spanish participated in Experiment 2. They were
undergraduate students following different majors at the University of the Basque Country in
Donostia, Spain. At the moment of being tested, they were all enrolled in the national Language
School (Escuela Oficial de Idiomas), where they attended regularly the final level courses of
English as a second language. As can be seen in Table 3, their mean proficiency level was
relatively high. They all reported having normal or corrected-to-normal vision.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
4;%
Table 3. Characteristics of the bilingual participants in Experiment 2 and 3.
Spanish-English Dutch-English Statistic p
Age 27.02 19.49 W* = 3686 0.00
Female/Male 39/27 44/21 X2 = 0.71 0.40
Age of acquisition 7.80 11.92 W = 581 0.00
Age of relative proficiency 19.05 16.55 W = 2819 0.00
Overall proficiency 7.35 7.34 W = 2089 0.79
Overall comfort 7.26 6.85 W = 2449 0.15
Speaking skill 7.21 7.08 W = 2207 0.77
Comprehension skill 7.57 8.30 W = 1443 0.00
Reading skill 7.92 7.58 W = 2614 0.03
Exposure (hours per week) 11.99 7.18 W = 2968 0.00
* Wilcoxon rank sum test with continuity correction
All other methodological aspects were kept identical to Experiment 1.
Results and Discussion
We cleaned and analyzed the data in the same way as was done for Experiment 1.
Additionally, we tested the interaction of Relatedness, Prime Type and L1 (English - Spanish) in
the joint results of Experiment 1 and 2. Trimming led to the removal of 14 individual RTs (0.2%
of the data). Condition means are shown in Table 4.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
4<%
Table 4. Mean reaction times and accuracy proportions per condition in Experiment 2.
Prime Type Relatedness RT Accuracy
Form related 703 0.88
unrelated 717 0.85
effect 14 -0.02
Opaque related 683 0.92
unrelated 708 0.90
effect 25 -0.02
Transparent related 654 0.97
unrelated 689 0.95
effect 35 -0.02
The RT analysis showed a significant interaction of Relatedness and Prime Type (F(2,
9894) = 5.13, p < .01). Priming was again largest in the transparent condition (35ms; t = 9.16, p <
.001) and smallest in the form condition (14ms; t = 4.65, p < .001; interaction: t = 3.19, p < .01).
The opaque condition showed an intermediate effect (25ms; t = 7.22, p < .01). The individual
comparisons with the transparent and form condition failed to reach significance, however (t =
1.38, p = .17 and t = 1.82, p = .09, respectively). The joint analysis of Experiment 1 and 2 showed
no interactions with L1. The individual effect of L1 was nevertheless significant (F(1, 19643) =
28.52, p < .001; the native participants were on average 71ms faster).
While the Relatedness x Prime Type interaction was not significant in the accuracy
analysis, the overall Relatedness effect was significant (z(9897) = 3.25, p < .01; 2% higher
accuracy following related primes). The joint analysis showed no interaction of this effect with
L1. There was an interaction between L1 and Prime Type (X2(2) = 37.57, p < .001), showing
significantly higher accuracies in Experiment 1 only for targets in the opaque and form condition
(4% more; z(19641) = 3.79, p < .001; and 7% more; z(19641) = 5.62, p < .001, respectively).
The statistical analysis indicates that even though participants were processing their
second language, a similar pattern to that of Experiment 1 arose: large facilitation for transparent
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
4=%
items, significantly smaller facilitation in the form condition and an intermediate effect for
opaque items. Although the transparent-opaque and opaque-form comparisons failed to reach
significance (p = .16 and p = .07), the numerical differences (10ms and 11ms) are again not
surprising, given the distributions in the recent meta-analyses (Rastle & Davis, 2008; Feldman et
al., 2009; Davis and Rastle, 2010). The results of Experiment 2 do not follow the findings of Silva
and Clahsen (2008) with –ness and –ity derivations, but instead indicate that non-native bilinguals
adopt the same strategy as native speakers in processing derivations.
If we compare the separate analyses of Experiment 1 and 2, it nevertheless appears that,
unlike natives, the Spanish-English bilinguals show clear form priming (14 vs. 1ms). At the same
time, the transparency effect seems somewhat smaller (i.e., 10ms instead of 21ms). This pattern is
in line with the conclusions of Feldman et al. (2009) in the context of inflections. Indeed, it
suggests a greater reliance on form processing in L2. This could result from slower prime
processing. In the model proposed by Taft and Nguyen-Hoan (2010), for instance, activations
flow from graphemic units to morpho-orthographic units, to lemma units (and back). When
primes are processed more #+'>+?, the relative contribution of graphemic overlap and morpho-
orthographic structure to target facilitation should become larger, while the contribution of
morpho-semantic structure becomes smaller. We need to remain cautious about these
interpretations, however, as the apparent differences with L1 are not supported statistically and
undoubtedly fall within the between-experiment variability observed in the L1 literature.
In Experiment 3, we test a group of Dutch-English participants with relatively high
English proficiency, allowing us to check the generality of our conclusions thus far. If bilinguals
adopt the same morphological processing strategy as natives, we should observe a similar pattern
as in Experiment 1. As can be seen in Table 1, the participants in Experiment 3 started learning
English at a later age and report lower reading skills and less exposure. Despite this, they report
an earlier age for relative proficiency and better comprehension skills (footnote 6). If significant
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
6@%
differences with Experiment 1 emerge, we will need to interpret them along these proficiency
characteristics.
Experiment 3
Method
Participants
Sixty-five native speakers of Dutch participated in Experiment 3. They were recruited
from the undergraduate student population of the faculty of Psychology and Educational Sciences
at Ghent University. All participants graduated from secondary school# where English is taught
mandatorily as a second language from the age of 12-13 onwards. Their proficiency
characteristics are listed in Table 3. They all reported having normal or corrected-to-normal
vision.
All other methodological aspects were kept identical to the previous experiments.
Results and Discussion
We cleaned and analyzed the data in the same way as was done for Experiment 2.
Trimming led to the removal of 3 individual RTs (0.03% of the data). Condition means are shown
in Table 5.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
64%
Table 5. Mean reaction times and accuracy proportions per condition in Experiment 3.
Prime Type Relatedness RT Accuracy
Form related 744 0.85
unrelated 758 0.86
effect 14 0.01
Opaque related 709 0.91
unrelated 735 0.88
effect 26 -0.03
Transparent related 699 0.94
unrelated 734 0.92
effect 35 -0.02
The Relatedness x Prime Type interaction was again significant for the RT data (F(2,
9744) = 7.35, p < .001). As before, priming was largest in the transparent condition (35ms; t =
8.63, p < .001), smallest in the form condition (14ms; t = 3.73, p < .001; interaction: t = 3.47, p <
.001) and intermediate in the opaque condition (26ms; t = 4.17, p < .001). The individual
comparison of the opaque effect with the transparent and form condition was only significant in
the former case, however (t = 3.15, p < .01 and t < 1, respectively). In the joint analysis of
Experiment 1 and 3, there were no interactions with L1. The individual effect of L1 was again
highly significant (F(1, 19493) = 38.10, p < .001; the native participants were on average 108ms
faster).
Unlike in the previous experiments, there was a significant interaction of Relatedness and
Prime Type in the accuracy analysis (X2(2) = 10.74, p < .01). There was 2% more accuracy
following related primes in the transparent condition (z(9742) = 2.59, p < .05) and 3% more in the
opaque condition (z(9742) = 3.51, p < .01). There was no significant difference between these
effects and both differed significantly from the form condition (z(9742) = 2.38, p < .05 and
z(9742) = 3.01, p < .01, respectively). The joint analysis with Experiment 1 showed no significant
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
66%
interactions with L1. The individual effect was nevertheless significant (z(19493) = 6.61, p <
.001; 6% more accuracy for natives).
Experiment 3 again indicates that (late) bilinguals adopt a morphological processing in
their L2 strategy similar to that of native speakers of that language. We replicate a graded priming
pattern, with largest facilitation for transparent items, smallest for form items and an intermediate
effect with opaque items. There were no significant differences with the outcome for natives,
again contrary to the findings of Silva and Clahsen (2008). The condition means are quasi-
identical to Experiment 2 (cf. Table 4 and 5), which implies that, compared to Experiment 1, form
priming is more evident and the semantic transparency effect is somewhat smaller. We again need
to remain cautious about this pattern, however. In the final part of our study we take advantage of
the similarity of the two bilingual data sets and merge them together in order to compare them
with the monolingual data. This analysis will determine our final conclusion about the influence
of L2 processing in masked morphological priming.
The joint analysis also serves to establish the graded nature of priming. Up to now, the
individual comparisons of the conditions did not always reach significance. Increasing statistical
power should help to bring clarity in this respect. Finally, to help our theoretical discussion, we
take advantage of the full power of our data to investigate the role of a number of important item
characteristics.
For our primes we entered the covariates Stem Family Size and Frequency, Suffix Family
Size and Frequency, Suffix Form Frequency, Boundary Frequency, Semantic Transparency,
Prime Neighborhood Size and Frequency and Word Form Frequency as measured in the related
condition. The stem family comprises all words that are derived from a given stem (e.g., view,
viewer, viewpoint, etc.), whereas the suffix family comprises all words that contain a given suffix
(viewer, baker, header, etc.). We calculated both the size and summed (lemma) frequencies of the
family members in each set. A positive effect of these (correlated) morpheme frequencies on the
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
67%
magnitude of priming serves as a general marker for specialized morphemic unit involvement.
The suffix form frequency represents the summed frequency of all word forms that carry the
suffix as an orthographic word ending (e.g., viewer, corner, later, pier, etc.). This measure probes
the reliance of priming on the presence of highly recurrent orthographic unit at the word ending.
The boundary frequency counts the frequency of the bigram at the (pseudo-) morpheme boundary
(e.g., viewer:we, corner:ne, freeze:ez). A negative effect on priming of this measure would
support the reliance on local sublexical regularities (cf. Seidenberg, 1987; Rastle et al., 2004).
LSA scores for the related prime-target pairs provide an opportunity to measure the effect of
semantic transparency continuously. Positive effects would confirm the conditioning of priming
by morpho-semantic characteristics. The lexical neighborhood was defined as all words located at
an orthographic edit distance of exactly one. Primes with large and/or high frequency
orthographic neighborhoods are potentially less effective due to an increased competition at the
level of lexical form representations. If the observed transparency effects indeed have a lexical
origin, we might especially observe such a negative effect for priming with transparent items.
Finally, the word form frequency of our related primes serves as a diagnostic for whole-word
processing. If transparency effects rely on such processing, this could manifest itself as a positive
correlation with priming for transparent items and potentially also a negative effect in the case of
opaque primes. The latter is not necessarily the case, since at least within some time window
morphological effects for opaque items and semantic transparency effects appear to co-exist. This
is in line with the idea that the lexical representation of the opaque suffixed prime does not
actively inhibit the lexical representation of the (pseudo-)stem (e.g., Taft & Nguyen-Hoan, 2010).
Apart from these prime characteristics, we investigated whether priming was affected by
the frequency and orthographic neighborhood characteristics of our targets (i.e., Target Word
Frequency, Target Neighborhood Size and Frequency). The frequency of a target clearly stands
out as the best predictor of lexical decision performance (e.g., Balota, Cortese, Sergent-Marshall,
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
68%
Spieler & Yap, 2004; Keuleers, Diependaele & Brysbaert, 2010). More frequent targets are
responded to faster, which potentially makes them less susceptible to a priming manipulation.
Hence, we might observe a general negative correlation between target frequency and facilitation.
The orthographic neighborhood size and frequency of our targets could also result in a negative
correlation with priming. The reason is that representations of targets are more difficult to activate
within large and/or high frequency neighborhoods, given a stronger amount of form-based
competition.
Joint Analysis
We merged the data of our 3 experiments. The factor L1 was now coded as native
(Experiment 1) vs. non-native (Experiment 2 & 3).
Results and Discussion
L1 x Relatedness x Prime Type
There were no interactions with L1 in both the RT and accuracy data. There was of course
a large individual effect; RTs were 89ms smaller for natives and accuracies 5% higher (F(1,
29393) = 49.01, p < .001 and (X2(1) = 37.86, p < .001)). Based on this outcome, our final
conclusion is that there is no qualitative difference in processing semantically transparent and
opaque complex words in English for L1 and L2 users. This does not exclude significant
differences at lower proficiency levels, but it does mean that eventually, with increasing skills,
these differences disappear.
Relatedness x Prime Type
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
69%
The interaction of Relatedness and Prime Type (F(2, 29393) = 19.45, p < .001) showed
that the transparent priming effect (36ms; t = 14.28, p < .001) was significantly larger than both
the opaque effect (22ms; t = 8.42, p < .001; interaction: t = 4.15, p < .001) and form effect (10ms;
t = 5,65 p < .001; interaction: t = 6.11, p < .001) and that the opaque effect was in turn
significantly larger than the form effect (t = 1.96, p < .05). We therefore conclude that we are
indeed dealing with a graded pattern of priming across our three conditions, reflecting
morphological effects with both transparent and opaque primes together with a positive influence
of semantic transparency.
The Relatedness x Prime Type interaction was also significant in the accuracy data (X2(2)
= 8.01, p < .05). There was a 2% higher accuracy following related primes in the transparent
condition (z(29391) = 4.12, p < .001) and 2% more in the opaque condition (z(29391) = 3.70, p <
.001). There was no significant difference between these effects and both differed significantly
from the form condition, although only marginally so for opaque items (z(29391) = 2.64, p < .05
and z(29391) = 1.96, p = .07, respectively). The absence of a transparency effect in the accuracy
data should be treated with caution. Overall, accuracies were quite high, making it difficult to
pick up condition differences.
Item characteristics
All covariates were log-transformed and centered to their mean (! = 9.65; Belsley, Kuh, &
Welsch, 1980). We conducted separate analyses for each condition (transparent, opaque, form).
The first reason is that the morpheme frequencies for suffixes are not available in the form
condition. A second reason is that some of the covariates had different values across the three
conditions (cf. Table 1). Suffix Form Frequency was lower in the form condition than in the
transparent and opaque condition. Semantic Transparency was larger in the transparent condition.
These differences are of course a natural consequence of our design. Two additional results were
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
6:%
that Stem Family Size was larger for transparent than for form items and Suffix Family Size and
Frequency were higher for transparent items. Considering the observed priming differences, these
5 unmatched variables are likely to interact with Relatedness in a joint analysis. We need to
ascertain, however, that such interactions are present at the level of our items and not (merely) at
the level of our conditions. Hence, we conduct three separate analyses on the joint data, where we
test interactions between our covariates and Relatedness for the transparent, opaque and form
condition, respectively. In each analysis, we entered covariates in interaction with Relatedness
following a stepwise forward selection procedure.
For the RT data in the transparent condition, Relatedness interacted significantly with
Suffix Form Frequency (t = 2.71, p < .01), Boundary Frequency (t = 2.45, p < .05), and Prime
Frequency (t = 2.22, p < .05). Whereas Suffix Form Frequency and Prime Frequency showed a
positive effect on priming, the effect of Boundary Frequency was negative. For opaque items,
only Target Neighborhood Frequency interacted with Relatedness (t = 2.82, p < .01). The sign of
this effect was negative: priming was smaller for targets with a high frequency neighborhood.
There were no significant interactions with Relatedness in the form condition. This was also the
case in each of the three accuracy analyses.
The results of our final analyses show an interesting different pattern of covariance across
our three priming conditions. Only in the transparent condition, priming increased with increasing
prime frequency. This supports the existence of positive lexical links between transparent
suffixed words (primes) and their stem (targets). Such links could be explicit, as in the models of
Taft and Nguyen-Hoan (2010) and Diependaele et al. (2005, 2009), for instance, but equally well
implicit. An example of the latter possibility is lateral inhibitory links between lexical items that
are less strong in the case of transparent morphological relatives. Our covariance analysis also
nicely illustrates the role of sublexical variables. The more frequent the suffix, as an orthographic
string, and the less frequent the morpheme boundary bigram, the more priming we observe for
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
6;%
transparent items. This clearly provides support for a morpho-orthographic system that detects
morphemes on the basis of the distinct distributional properties of letters and/or phonemes within
morphologically complex words (cf. Rastle et al., 2004). The fact that this effect was only
observed in the transparent condition could be related to the presence of a negative component in
the opaque condition, i.e., an inhibitory target neighborhood frequency effect. The latter effect
shows that although sublexical features give rise to morphological effects with opaque primes,
these effects nevertheless show traces of lexical processing. It thus appears that morpho-
orthographic activations are rapidly manifested at lexical levels. A final point regarding the
covariance results is that some readers might be tempted to interpret the absence of effects of
stem family size and continuous semantic transparency (LSA) as evidence against any semantic
or morpho-semantic influence. Null-effects can never be taken as “negative evidence”, however.
Diependaele et al. (2009), for instance, did report a significant correlation of masked
morphological priming with semantic transparency. They arguably used a more fine-grained
measure, based on a large-scale rating study.
General Discussion
The present study investigated the processing of suffix-derivations in first and second
language visual word recognition. We compared masked morphological priming in English across
a group of native English, Spanish and Dutch speaking participants. The design followed that of
many recent monolingual studies, including stem priming with semantically transparent suffixed
primes (e.g., viewer-view), semantically opaque (including pseudo-) suffixed primes (e.g., corner-
corn) and stem-embedded form control primes (e.g., freeze-free). Following the work of H.
Clahsen, we were specifically interested in whether the results would show that bilinguals who
reach L2 proficiency relatively late make less use of morphological information in the processing
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
6<%
of suffixed derivations. Contrary to Clahsen et al.’s conclusions, similar priming patterns emerged
for the native English participants and the two groups of bilinguals. The actual pattern we observe
fits well with recent meta-analyses of the L1 masked morphological priming literature (Feldman
et al., 2009; Davis & Rastle, 2010). Priming was largest with transparent primes, smallest with
form primes and intermediate with opaque primes.
Native and non-native morphological processing
Our results diverge from Silva and Clahsen (2008), who found evidence for reduced
morphological priming with suffix derivations in L2. We believe this is most likely due to a
number of methodological shortcomings. First, Silva and Clahsen only used a limited number of
items. Throughout the study, there were only 6 to 7 items per condition. They also only tested
derivations with the suffixes -ness and -ity. Such limitations clearly allow questioning the
generality of their findings. In the present experiments there were 25 items per condition with a
range of different derivational suffixes. A further shortcoming is that none of the Clahsen et al.
studies took measures to deal with the high error rates (i.e., missing data) that are typically
present for bilingual participants. Lexical decision errors are far from random, which excludes
simple mean replacement as a valuable technique to deal with missing values (footnote 7). In the
present study we used state of the art techniques to deal with this (general) problem in L2
research (cf. footnote 3). It is also important to note that the present conclusion that there exist no
qualitative or quantitative differences in the processing of suffixed derivations in L1 and L2 does
not exclude that such differences do arise at earlier stages in L2 acquisition. The difference with
Silva and Clahsen (2008) could also be considered from this perspective. Indeed, while our
participants started to learn English at about 8 and 12 (and possibly much earlier via audio-visual
media), the participants in the Clahsen et al. studies were only initially exposed to their L2
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
6=%
(English/German) at the age of 14 (on average). Furthermore, an account where L2 processing
gradually approaches the functional architecture for natives can also predict that the speed by
which this occurs is a positive function of the linguistic distance between an individual’s L1 and
L2. In this respect, it is potentially important that Silva and Clahsen’s study included Chinese-
English and Japanese-English participants. Over and above the script change between those
language combinations, it is noteworthy that the distance between Spanish and English and Dutch
and English is considerably smaller (Chiswick & Miller, 2004; Serva & Petroni, 2008).
It can be argued that the present conclusion, i.e., that morphology plays a similar role in
the processing of suffix derivations in L1 and L2, has in fact nothing to say about the processing
of inflections in L2. Indeed, comparing the present data with those of Neubauer and Clahsen
(2009), for instance, one could try to explain the difference via a model where derivations and
inflections are treated in a fundamentally different way (e.g., Clahsen et al., 2003). Recall that
Neubauer and Clahsen (2009) found that stem priming with regular past participles (worked-
work) was eliminated in L2, while priming with irregular forms remained stable (taught-teach).
According to Clahsen et al. (2003), in L1 derivations as well as irregular inflections are stored as
full forms in memory with reference to their morphological constituents. Regular inflections in
L1, on the contrary, are not stored. They are produced and recognized by applying an online
grammatical rule. This online computation of inflected forms and the retrieval of all other forms
lie at the heart of the so-called dual-mechanism theory (e.g. Pinker, 1999). According to this
view, the present findings have nothing to say about the domain of regular inflection. The
declarative/procedural model (Ullman, 2004, 2005) states that L2 word recognition primarily
relies on the direct lexical retrieval route in the dual-mechanism view, and not on the online
segmentation route. It is precisely the direct route where morphological effects with derivations
arise according to Clahsen et al. (2003). Hence, it is not surprising that we observe similar effects
in L1 and L2, while Neubauer and Clahsen (2009) find a marked difference for regular
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
7@%
inflections. Although this line of reasoning can be followed for the transparent items in our study,
it arguably cannot for the morphological effects with opaque items. Our results with opaque items
specifically show the same evidence for morpho-orthographic processing in L2 and L1. Such
processing is typically conceptualized as an online sublexical morphological segmentation
mechanism that applies to both derivations and regular inflections. Within the above line of
reasoning it would thus appear that while the online morphological segmentation of regular
inflections is heavily reduced and potentially nonexistent in L2 processing, it remains unaffected
in the case of derivations. In other words, to explain the present results and those of Neubauer and
Clahsen (2009) within the dual-mechanism view, one needs to assume that the sublexical
segmentation of regular inflections and derivations occurs via separate mechanisms, which are
also differentially affected by maturation. To our knowledge, there is no a priori reason for such
an assumption. Therefore, it seems that the present data limit the scope of the
declarative/procedural approach to L1 and L2 processing and its projection onto dual-mechanism
theory. Future research needs to address this issue thoroughly. It is important to note that, in
research on morphological productivity, single-mechanism computational models have been very
successful at generating patterns of results that were regarded as evidence for a dual mechanism
(e.g., Albright & Hayes, 2003; Chandler, 2010; Hahn & Nakisa, 2000; Keuleers et al., 2007;
Keuleers & Daelemans, 2007). Recall also that the priming results of Clahsen et al. with respect
to inflections have already been challenged by Feldman et al. (2009).
Our general conclusion that the processing of derivations in a given language takes on a
similar functional shape, regardless of nativeness is in fact not an isolated finding in the literature.
It is also supported by the work of Frost, Kugler, Deutsch and Forster (2005). With both Hebrew-
English as well as English-Hebrew bilinguals, these authors found reliable masked form priming
(e.g., cat-hat) in English, but not in Hebrew. Instead, the Hebrew results showed masked
morphological priming, irrespective of semantic transparency. This pattern (i.e., form and
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
74%
morphological priming in English versus mere morphological priming in Hebrew) is fully
compatible with the monolingual studies in both languages. Frost et al. conclude that the form
similarities that define lexical competition among words are primarily determined by language-
specific distributional properties. It is important to note that such a conclusion does not imply that
L1 and L2 lexical representations have a different functional location. Indeed, many empirical
results support that L1 and L2 processing occurs within the same architectural environment. This
explains, for instance, why bilinguals are faster to recognize words whose form and meaning are
highly similar across the languages they know (e.g., appel-apple for Dutch-English bilinguals;
e.g., Dijkstra, Grainger & Van Heuven, 1999; Duñabeitia, Perea, & Carreiras, 2010; Van Hell &
Dijkstra, 2002; Van Assche, Duyck, Hartsuiker & Diependaele, 2009). Furthermore, there is
ample evidence that orthographic and phonological encoding occurs language-independently and
that the results interact with each other (e.g., Van Heuven, Schriefers, Dijkstra & Hagoort, 2008;
Dimitropoulou, Duñabeitia & Carreiras, 2010). Known languages are thus not represented, nor
processed independently from each other. Several models of bilingual word processing
incorporate this idea. The Bilingual Interactive Activation model (BIA; Dijkstra, Van Heuven &
Grainger, 1998; Van Heuven, Dijkstra, & Grainger, 1998) and its more recent version, the BIA+
(Dijkstra & Van Heuven, 2002), for instance, assume that an input word simultaneously activates
many related words in a language-independent manner. The present results simply imply that
despite the integrated nature, L1 and L2 processing dynamics evolve to a state wherein they are
primarily determined by their own linguistic properties. The exact same conclusion was drawn by
Lemhöfer, Dijkstra, Schriefers, Baayen, Grainger and Zwitserlood (2008). They conducted a
large-scale regression study where English, Dutch, French and German participants completed the
same English word recognition task. The results showed that only predictors specific to English
language accounted for significant proportions of the variance, the only exception being cognate
status.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
76%
The role of form and meaning in morphological processing
The graded nature of the priming pattern in our data, where priming correlated with
morphological (stem+suffix) surface structure and semantic transparency, further demonstrates
that fast morphological activation is conditioned by both form and meaning characteristics. This
can be accounted for in different ways, as illustrated in Figure 1. According to the first general
architecture, there is a mandatory sublexical segmentation stage leading to the activation of
morphemic representations regardless of semantic transparency. Positive links between the lexical
representations of transparent relatives result in stronger morphemic activations for transparent
derivations later on during processing. There are many possibilities for conceptualizing these
links explicitly or implicitly. As we already discussed in the introduction, this view can account
for the simultaneous observations of semantic transparency effects and morphological effects for
opaque items with short prime durations if the assumption is made that there is a (relatively)
passive decay of morpho-orthographic activations when they are not further supported (e.g., Taft
& Nguyen-Hoan, 2010). In principle, this first approach predicts a moment in time where
morphemic activations are purely morpho-orthographic in nature (i.e., equivalent across different
levels of semantic transparency). This prediction is potentially very hard to verify empirically,
however. If one assumes cascaded processing, incoming information (i.e., activation) does not
have to be fully analyzed at one level before being passed on to more abstract levels. A lexical
morphological influence could thus already start to develop before the representations involved in
morpho-orthographic processing have reached a critical activation level needed for observing
facilitation in the masked priming paradigm. Potentially, measures with a higher time-resolution,
such as event-related potentials, and/or a higher activity-resolution, such as functional magnetic
resonance, provide better candidates for testing this prediction (see Devlin, Jamison, Matthews &
Gonnerman, 2004; Lavric, Clapp & Rastle, 2007; Morris, Holcomb & Grainger, 2008). The
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
77%
presence of semantic transparency effects in our study is thus not necessarily problematic for this
account. It can also be remarked that the prime duration, 53ms, is slightly higher than the central
tendency in the literature (median = 48, mean = 46, Q1 = 42, Q3 = 52, sd = 7; Rastle & Davis,
2008, Table 1). It could be argued that this contributed to the presence of semantic transparency
effects, providing more time for lexical activations to become influenced by the primes.
Figure 1. Three architectures that can account for the graded pattern of masked priming across
the transparent, opaque and form conditions. The general framework in panel A predicts a
mandatory sublexical morphological decomposition (e.g., Rastle and Davis, 2008; Taft &
Forster, 1975; Taft & Nguyen-Hoan, 2010). Once the results of this decomposition are translated
onto lexical levels, morpho-semantic relationships can come in to play. This happens because of
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
78%
explicit or implicit positive links between representations of transparent relatives. Panel B shows
a parallel dual-route architecture with simultaneous morpho-orthographic and morpho-semantic
processing (e.g., Diependaele et al., 2005, 2009). The critical difference with the former models is
that there is a direct link between low-level sublexical representations and lexical
representations. This route in principle allows observing pure morpho-semantic effects, i.e.,
morphological effects without the involvement of a sublexical decomposition mechanism. In
practice, except for irregular forms, there is cooperation between the direct and the sublexical
decomposition route. Panel C illustrates the distributed connectionist view on morphological
processing (e.g., Plaut & Gonnerman, 2000). Through statistical learning, continuous hidden unit
representations (depicted as bar plot patterns) capture systematic correlations between form and
meaning. Since morphological families are prototypical for such correlations, the representations
for transparent relatives become more similar to each other than the representations for words
that are only related in form or meaning.
According to the two other accounts depicted in Figure 1, semantic transparency plays a
central role in the emergence of morphological activations and the present results in L1 and L2
can be viewed as a further illustration of this. The first model can be labeled as a parallel dual-
route model with morpho-orthographic and morpho-semantic processing. The main departure
from the previous architecture, is that there is a parallel and simultaneously operating route that
provides the possibility of a fast direct activation of a lexical representations matching the whole
input. The presence of positive links between lexical representations of transparent morphological
relatives implies that this direct whole-word route can in principle generate masked
morphological priming effects on its own, which is referred to as morpho-semantic processing.
However, since morpho-orthographic segmentation is initiated at the same time, in practice there
will be cooperative effects of the two sources of morphological activation, except for irregular
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
79%
forms (e.g., Diependaele et al., 2005, 2009). In this view, semantic transparency is able to affect
morphological activations from the very start, given that the direct whole-word route is assumed
to provide the fastest activation of lexical representations (see also Grainger & Dufau, 2010;
Schreuder & Baayen, 1995).
The final account of the observed graded pattern has been formulated within the
distributed connectionist approach to language processing (e.g., Harm & Seidenberg, 2004). In
this view representations of form and meaning are in essence viewed as continuous values across
a given set of features, i.e., orthographic/phonological and semantic vectors of a given
dimensionality. The mappings between these representations are in turn viewed as continuous
representations at the level of an intermediate (“hidden”) feature set. Importantly, the values of
these representations are the result of statistical learning. As a general rule, the more frequent and
consistent a given form meaning feature mapping occurs, the more similar/or closer
together/overlapping the hidden feature values will be across the individual occurrences. In the
context of morphology this predicts that the hidden representation of morphological relatives will
have a relatively high degree of similarity and that this is a positive function of the number of
semantically transparent family members and their frequency/probability distribution, i.e., the
higher the overall frequency/probability and the more equated across family members, the more
similar hidden representations will be. Priming effects are assumed to provide a direct reflection
of the similarity. Facilitation will thus generally be larger following morphologically related
primes than other matched controls. Simulations with artificial language by Plaut and Gonnerman
(2000) illustrate that at least under certain circumstances one can expect that hidden
representations for opaque words exhibit some degree of similarity to those of the genuine (i.e.,
transparent) morphological family. As such, it is possible to predict the graded patterns of priming
across transparent, opaque and form items shown in the literature.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
7:%
The present findings do not provide a strong basis for deciding between these alternatives.
Our covariance analyses nevertheless provide interesting information in this regard. First, there
was a positive relationship between the facilitation for transparent items and the whole-word
prime frequency. This is readily predicted by the first two accounts, where positive links are
present between lexical representations of transparent morphological relatives. In the distributed
connectionist account it is especially the family size and frequency of the stem that should
positively influence priming. In fact, the word frequency of the prime could be predicted to have a
negative effect on priming, since it would support a more idiosyncratic hidden representation of
the prime word. The negative effect of the morpheme boundary bigram frequency in the case of
transparent items is also easily integrated with the former two proposals. As shown by Rastle et
al. (2004), the differential bigram frequency contours in morphologically complex and simplex
words could play a crucial role in morpho-orthographic segmentation. It is less straightforward to
see how the observed negative effect is to be explained within the distributed connectionist view.
Arguably, such dependencies can only arise from mappings between different levels of
orthographic/phonological representation and not from the mapping of form onto meaning.
Finally, our covariance analyses also show a negative effect of stem neighborhood frequency in
the case of opaque items. This can be accounted for within the two former frameworks if one
assumes that morpho-orthographic activations are rapidly translated onto lexical levels. The more
frequent the stem orthographic neighborhood, the harder it will be to pre-activate its lexical
representation due to stronger competition from words with a similar form. In the case of
transparent items, the positive links within the morphological family could immediately neutralize
this competition. Again, it remains to be seen whether a distributed connectionist model can
capture this effect.
Following the above discussion it would seem that our results are most easily explained by
a pure morpho-orthographic and a simultaneous morpho-orthographic and morpho-semantic
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
7;%
model. The former account is arguably more parsimonious, but it remains important to study its
ability to predict large effects of semantic transparency and whole-word processing in general at
an early stage. The masked priming studies that have led to the revival of this “blind
decomposition” framework in the last decade, all showed statistically equivalent priming effects
with transparent and opaque items (e.g., Longtin et al., 2003; Rastle et al., 2004). Current
evidence, including the present data, clearly shows that fast semantic transparency effects cannot
be overlooked. If we consider the recent meta-analysis by Davis and Rastle (2010), it appears that
our joint analysis enters the literature as the one with the largest number of data points for the
transparent-opaque and opaque-form priming contrasts. Following their calculations, we have
more than 3 times the maximum number thus far (i.e., 9800 as opposed to 3000). This means that
the present grand averages for the transparent-opaque and opaque form contrasts (13ms and
12ms) would appear well on top in the funnel plots. Especially if the means across studies were
weighted by the number of data points, it would seem that the Davis and Rastle analysis
underestimates the transparent-opaque difference (7ms vs. 14ms in the present study) and
overestimates the opaque-form difference (20ms vs. 12ms). Thus, although a pure morpho-
orthographic explanation can account for a graded priming pattern as presently observed,
proponents should be aware that semantic transparency effects in fast morphological priming are
very real and potentially more equal in size to the morphological effect with opaque items than
the present meta-analyses suggest.
A simultaneous morpho-orthographic/semantic account obviously goes hand in hand with
accepting the reality of fast semantic transparency effects. The critical difference is the
independent origin of morphological effects with opaque items and semantic transparency effects.
Like earlier parallel dual-route accounts (e.g., Schreuder & Baayen, 1995), the speed of whole-
word processing (leading to morpho-semantic activation) and “blind” decomposition (morpho-
orthographic activation) is relative, rather than absolute. This is a critical property that needs to be
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
7<%
considered in future research. A further reason for considering this alternative is that it readily
accounts for masked priming effects with irregular inflections. Indeed, in a recent set of masked
priming lexical decision experiments, Crepaldi, Rastle, Colthaert and Nickels (2010) found that
irregularly inflected primes (fell) consistently primed their stems (fall) relative to both
orthographic (e.g., full-fall) and morpho-orthographic (raid-ray, cheese-choose) controls.
Crepaldi et al. argue that this can be accounted for by extending the pure morpho-orthographic
view with a lemma level. This level contains shared representations for inflectional, but not
derivational morphological relatives. They argue against the possibility of a morpho-semantic
level, i.e., a level where representations instead reflect the combination of orthographic and
semantic similarity for all words, because “masked priming studies typically show that priming
effects for semantically transparent derivational pairs like darkness-DARK do not differ
significantly from priming effects for pseudo-morphological pairs like corner-CORN (see Longtin
et al. (2003), Marslen-Wilson, Bozi! & Randall (2008), Rastle et al. (2004)). If the combination
of semantic and orthographic similarity were playing a strong role in masked priming, then it
seems that a convincing effect should be apparent across this comparison.” (Crepaldi et al., 2010,
p. 91). The present conclusion regarding semantic transparency effects clearly sheds a different
light on this line of reasoning. Fast semantic transparency effects are not to be overlooked and
simultaneous morpho-orthographic/-semantic processing readily predicts that irregular
inflectional primes will only result in a morpho-semantic effect. This effect will also be
considerably larger than the semantic transparency effect with derivations, given the generally
stronger semantic similarity among inflections.
The question remains, however, why there exists such considerable variability regarding
masked priming with transparent and opaque items. We believe at least one contributing factor
might be that a factorial treatment of semantic transparency can be misleading, given the nature of
this variable. There is clearly an underlying continuum with a considerable grey area when it
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
7=%
comes down to building a factorial contrast. Our covariance results further highlight a number of
potentially important variables. We can arrive at the prediction that priming will become larger
for transparent, but not opaque items, the higher the prime word frequency is. At the same, we
can predict that priming with opaque will become smaller as the target neighborhood frequency
becomes larger. As such, studies with relatively high frequency primes and high frequency target
neighborhoods should have a better chance of providing semantic transparency effects or, put
differently, studies with relatively low frequency primes and low frequency target neighborhoods
should have a better chance of observing matched facilitations across transparent and opaque
items. Interestingly, compared to Rastle et al. (2004), the present target neighborhood sizes were
about 3-4 times as high. Following the above reasoning, this could explain why Rastle et al. found
a larger priming effect for opaque items than we did in Experiment 1 (i.e., 22ms vs. 15ms) and
why they failed to obtain a significant transparency effect. These hypotheses should be addressed
systematically in future research.
Conclusions
The present results show that the processing of morphologically complex words occurs
along similar principles and to the same degree in L1 and L2 processing. This supports the
general idea that the language itself primarily determines the functional properties of processing
in L2. This conclusion does not exclude that significant differences arise with lower levels of
proficiency. Future research should consider this possibility. If differences emerge, we argue that
this will reflect an intermediate state in the transition towards the target (L1) architecture rather
than a fundamentally different way of handling native and non-native language input. Regarding
the functional properties of morphological processing itself, our results highlight the reality of
semantic transparency effects within the very first stages of word recognition. We sincerely hope
that the scale of our study will be an incentive for future researchers to investigate the true
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
8@%
functional origin of these effects, rather than to look for presumed methodological flaws in
studies that show them.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
84%
References
Albright, A., & Hayes, B. (2003). Rules vs. analogy in English past tenses: A
computational/experimental study. Cognition, 90, 119–161.
Baayen, R. H. (2009). languageR: Data Sets and Functions with "Analyzing Linguistic Data: A
practical introduction to statistics". R package version 0.955. http://CRAN.R-
project.org/package=languageR.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-Effects Modeling with Crossed
Random Effects for Subjects and Items. Journal of Memory and Language, 59, 387-556.
Baayen, R. H., Piepenbrock, R., & Gulikers, L. (1995). The CELEX Lexical Database (CD-
ROM). Philadelphia: Linguistic Data Consortium, University of Pennsylvania.
Balota, D. A., Cortese, M. J., Sergent-Marshall, S. D., Spieler, D. H., & Yap, M. J. (2004). Visual
Word Recognition of Single-Syllable Words. Journal of Experimental Psychology:
General, 133, 283–316.
Bates, D., & Maechler, M. (2009). lme4: Linear Mixed-Effects Models Using S4 Classes. R
package version 0.999375-32. http://CRAN.R-project.org/package=lme4.
Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Regression Diagnostics: Identifying Influential
Data and Sources of Collinearity. New York: Wiley.
Benjamini, Y. and Hochberg, Y. (1995). Controlling the False Discovery Rate: A Practical and
Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society, Series B,
57, 289-300.
Bybee, J. L. (1985) Morphology: A Study of the Relation Between Meaning and Form,
Philadelphia: John Benjamins.
Chandler, S. (2010). The English past tense: Analogy redux. Cognitive Linguistics, 21, 371–417.
Chateau, D., Knudsen, E. V., & Jared, D. (2002). Masked Priming of Prefixes and the Influence
of Spelling-Meaning Consistency. Brain and Language, 81, 587–600.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
86%
Chiswick, B. R. & Miller, P. W. (2004). Linguistic Distance: A Quantitative Measure of the
Distance Between English and Other Languages. IZA Discussion Papers 1246, Institute for
the Study of Labor (IZA).
Clahsen, H., Felser, C., Neubauer, K., Sato, M. & Silva, R. (2010). Morphological Structure in
Native and Non-Native Language Processing. Language Learning, 60, 21-43.
Clahsen, H., Sonnenstuhl, I., & Blevins, J. P. (2003). Derivational Morphology in the German
Mental Lexicon: A Dual-Mechanism Account. In H. Baayen & R. Schreuder (Eds.),
Morphological Structure in Language Processing. Berlin: Mouton de Gruyter, 125-155.
Crepaldi, D., Rastle, K., Coltheart, M., & Nickels, L. (2010). ‘Fell’ Primes ‘Fall’, but Does ‘Bell’
Prime ‘Ball’? Masked Priming with Irregularly-inflected primes. Journal of Memory and
Language, 63, 83-99.
Davis, M. H. & Rastle, K. (2010). Form and Meaning in Early Morphological Processing:
Comment on Feldman, O’Connor and Moscoso del Prado Martin. Psychonomic Bulletin &
Review.
Devlin, J. T., Jamison, H. L., Matthews, P. M. & Gonnerman, L. M. (2004). Morphology and the
Internal Structure of Words. Proceedings of the National Academy of Sciences, 101(41),
14984-14988.
Diependaele, K., Grainger, J., & Sandra, D. (in press). Derivational Morphology and Skilled
Reading: An Empirical Overview. In M. Spivey, M. Joanisse & K. McRae (Eds.), The
Cambridge Handbook of Psycholinguistics, Cambridge UK: Cambridge University Press.
Diependaele, K., Sandra, D., & Grainger, J. (2005). Masked Cross-Modal Morphological
Priming: Unraveling Morpho-Orthographic and Morpho-Semantic Influences in Early
Word Recognition. Language and Cognitive Processes, 20, 75– 114.
Diependaele, K., Sandra, D., & Grainger, J. (2009). Semantic Transparency and Masked
Morphological Priming: The Case of Prefixed Words. Memory & Cognition, 37, 895-908.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
87%
Dijkstra T., Van Heuven W.J.B. (2002). The Architecture of the Bilingual Word Recognition
System: From Identification to Decision. Bilingualism: Language and Cognition, 5, 175-
197.
Dijkstra, T., Grainger, J., & Van Heuven, W.J.B. (1999). Recognition of Cognates and
Interlingual Homographs: The Neglected Role of Phonology. Journal of Memory and
Language, 41, 496-518.
Dijkstra, T., Van Heuven, W.J.B., & Grainger, J. (1998). Simulating Cross-Language
Competition with the Bilingual Interactive Activation Model. Psychologica Belgica, 38,
177-196.
Dimitropoulou, M., Duñabeitia, J.A., Carreiras, M. (in press). Phonology by Itself: Masked
Phonological Priming Effects With and Without Orthographic Overlap. European Journal
of Cognitive Psychology.
Duñabeitia, J.A., Perea, M., & Carreiras, M. (2008). Does Darkness Lead to Happiness? Masked
Suffix Priming Effects. Language and Cognitive Processes, 23, 1002-1020.
Duñabeitia, J.A., Perea, M., & Carreiras, M. (2010). Masked translation priming effects with
highly proficient simultaneous bilinguals. Experimental Psychology, 57(2), 98-107.
Feldman, L. B., Kosti!, A., Basnight-Brown, D. M., Filipovi!-"ur#evi!, D., & Pastizzo M. J.
(2009). Morphological Facilitation for Regular and Irregular Verb Formations in Native and
Non-Native Speakers: Little Evidence for Two Distinct Mechanisms. Bilingualism:
Language and Cognition, 13, 119-135.
Feldman, L. B., O'Connor, P. A., & Moscoso del Prado Martín, F. (2009). Early Morphological
Processing is Morpho-Semantic and not Simply Morpho-Orthographic: A Violation of
Form-then-Meaning Accounts of Word Recognition. Psychonomic Bulletin & Review, 16,
684-691.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
88%
Feldman; L. B. (1994). Beyond Orthography and Phonology: Differences between Inflections and
Derivations. Journal of Memory and Language, 33, 442-470.
Forster, K. I., & Davis, C. (1984). Repetition Priming and Frequency Attenuation in Lexical
Access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 680-
698.
Forster, K. I., & Forster, J. (2003). DMDX: A Windows Display Program with Millisecond
Accuracy. Behavioral Research Methods, Instruments & Computers, 35, 116-124.
Frost, R., Kugler, T., Deutsch, A., & Forster, K. I. (2005). Orthographic Structure Versus
Morphological Structure: Principles of Lexical Organization in a Given Language. Journal
of Experimental Psychology: Learning, Memory and Cognition, 31, 1293-1326.
Gelman, A., Hill, J., Yajima, M., Su Y.-S., & Pittau, M. G. (2010). mi: Missing Data Imputation
and Model Checking. R package version 0.08-06. http://CRAN.R-project.org/package=mi.
Giraudo, H., & Grainger, J. (2000). Effects of Prime Word Frequency and Cumulative Root
Frequency in Masked Morphological Priming. Language and Cognitive Processes, 15, 421-
444.
Grainger, J. & Dufau, S. (in press). The front-end of visual word recognition. To appear in J.S.
Adelman (Ed.) Visual Word Recognition Vol. 1: Models and Methods, Orthography and
Phonology. Hove, UK: Psychology Press.
Hahn, U., & Nakisa, R.C. (2000). German Inflection: Single or Dual Route? Cognitive
Psychology, 41, 313-360.
Harm, M. W., & Seidenberg, M. S. (2004). Computing the meanings of words in reading:
Cooperative division of labor between visual and phonological processes. Psychological
Review, 111, 662-720.
Keuleers, E., & Daelemans, W. (2007). Memory-based learning models of inflectional
morphology: a methodological case study. Lingue e linguaggio, 6, 151–174.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
89%
Keuleers, E., Diependaele, K. & Brysbaert, M. (in press). Practice Effects in Large-Scale Visual
Word Recognition Studies: A Lexical Decision Study on 14,000 Dutch Mono- and
Disyllabic Words and Nonwords. Frontiers in Language Sciences.
Keuleers, K., Sandra, D., Daelemans, W, Gillis, S., Durieux, G., & Martens, E. (2007). Dutch
Plural Inflection: The Exception that Proves the Analogy. Cognitive Psychology, 54, 283-
318.
Landauer, T.K. & Dumais, S.T. (1997). A Solution to Plato’s Problem: The Latent Semantic
Analysis Theory of Acquisition, Induction, and Representation of Knowledge.
Psychological Review, 104, 211-240.
Lavric, A., Clapp, A., & Rastle, K. (2007). ERP Evidence for Morphological Analysis from
Orthography: A Masked Priming Study. Journal of Cognitive Neuroscience, 19, 866-877.
Lemhöfer, K., Dijkstra, T., Schriefers, H., Baayen, H.R., Grainger, J., & Zwitserlood, P. (2008).
Native Language Influences on Word Recognition in a Second Language: A Mega-Study.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 34, 12-31.
Longtin, C. M., & Meunier, F. (2005). Morphological Decomposition in Early Visual Word
Processing. Journal of Memory and Language, 53, 26-41.
Longtin, C. M., Segui, J., & Hallé, P. A. (2003). Morphological Priming without Morphological
Relationship. Language and Cognitive Processes, 18, 313-334.
Marslen-Wilson, W. D, Bozi!, M. & Randall, B. (2008). Early Decomposition in Visual Word
Recognition: Dissociating Morphology, Form, and Meaning. Language and Cognitive
Processes, 23, 394-421.
Morris, J., Franck, T., Grainger, J., & Holcomb, P.J. (2007). Semantic Transparency and Masked
Morphological Priming: An ERP Investigation. Psychophysiology, 44, 506-521.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
8:%
Morris, J., Holcomb, P. J., & Grainger, J. (2008). An Electrophysiological Investigation of Early
Effects of Masked Morphological Priming. Language & Cognitive Processes, 23, 1021-
1056.
Morris, J., Porter, J. H., Grainger, J. & Holcomb, P. J. (in press). Effects Of Lexical Status And
Morphological Complexity In Masked Priming: An ERP Study. Language and Cognitive
Processes.
Neubauer, K. & Clahsen, H. (2009). Decomposition of Inflected Words in a Second Language:
An Experimental Study of German Participles. Studies in Second Language Acquisition, 31,
403-435.
Pinker, S. (1999). Words and rules: The ingredients of language. New York: Basic Books.
Plaut, D. C. & Gonnerman, L. M. (2000). Are Non-Semantic Morphological Effects Incompatible
with a Distributed Connectionist Approach to Lexical Processing? Language and Cognitive
Processes, 15, 445-485.
R Development Core Team (2009). R: A Language and Environment for Statistical Computing. R
Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL
http://www.R-project.org.
Rastle, K., & Davis, M. H. (2008). Morphological Decomposition Based on the Analysis of
Orthography. Language and Cognitive Processes, 23, 942-971.
Rastle, K., Davis, M. H., & New, B. (2004). The Broth in my Brother’s Brothel: Morpho-
Orthographic Segmentation in Visual Word Recognition. Psychonomic Bulletin & Review,
11, 1090–1098.
Raveh, M., & Rueckl, J. G. (2000). Equivalent Effects of Inflected and Derived Primes: Long-
Term Morphological Priming in Fragment Completion and Lexical Decision. Journal of
Language and Memory, 42, 103-119.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
8;%
Schafer, J. L., & Graham, J. W. (2002). Missing Data: Our View of the State of the Art.
Psychological Methods, 7, 147-177.
Schreuder, R., & Baayen, R. H. (1995). Modeling Morphological Processing. In Feldman, L. B.
(Ed.), Morphological aspects of language processing (pp. 131- 156). Hillsdale, NJ:
Erlbaum.
Schriefers, H., Frederici, A., & Graetz, P. (1992). Inflectional and Derivational Morphology in the
Mental Lexicon: Symmetries and Asymmetries in Repetition Priming. Quarterly Journal of
Experimental Psychology, 44, 373–90.
Seidenberg, M. S. (1987). Sublexical structures in visual word recognition: Access units or
orthographic redundancy? In Coltheart, M. (Ed.), Attention and Performance XII: The
Psychology of Reading (pp. 245-263). London: Erlbaum.
Serva, M., & Petroni, F. (2008). Indo-European Languages Tree by Levenshtein Distance. EPL,
81.
Silva, R., & Clahsen, H. (2008). Morphologically Complex Words in L1 and L2 Processing:
Evidence from Masked Priming Experiments in English. Bilingualism: Language and
Cognition, 11, 245-260.
Taft, M., & Forster, K. I. (1975). Lexical Storage and Retrieval of Prefixed Words. Journal of
Verbal Learning and Verbal Behavior, 14, 638-647.
Taft, M., & Nguyen-Hoan, M. (2010). A sticky stick: The locus of morphological representation
in the lexicon. Language and Cognitive Processes, 25, 277-296.
Ullman, M. (2004). Contributions of Memory Circuits to Language: The Declarative/Procedural
Model. Cognition, 92, 231-270.
Ullman, M. (2005). A Cognitive Neuroscience Perspective on Second Language Acquisition: The
Declarative/Procedural Model. In C. Sanz (Ed.), Mind and Context in Adult Second
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
8<%
Language Acquisition: Methods, Theory and Practice (pp. 141- 178). Washington, D.C.:
Georgetown University Press.
Van Assche, E., Duyck, W., Hartsuiker, R. & Diependaele, K. (2009). Does Bilingualism Change
Native-Language Reading? Cognate Effects in a Sentence Context. Psychological Science,
20, 923-927.
Van Hell J.G., Dijkstra T. (2002). Foreign Language Knowledge can Influence Native Language
Performance in Exclusively Native Contexts. Psychonomic Bulletin & Review, 9, 780-789.
Van Heuven, W., Schriefers, H., Dijkstra, A., & Hagoort, P. (2008). Language Conflict in the
Bilingual Brain. Cerebral Cortex, 18, 2706-2716.
Van Heuven, W.J.B, Dijkstra, T., & Grainger, J. (1998). Orthographic Neighborhood Effects in
Bilingual Word Recognition. Journal of Memory and Language, 39, 458-483.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
8=%
Author Note
This research was supported by grants PSI2009-08889 and CSD2008-00048 from the Spanish
Government. The authors would like to thank Raphael Serota and Maria Dimitropoulou for their
help in the data collection. They also thank Maggie Lynn for her help in revising the final
manuscript.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
9@%
Footnotes
1. Words like department are called semantically opaque because they originated as
true semantic derivatives of their stem, but gradually acquired an idiosyncratic meaning.
Words like corner are called pseudo-complex because they have no true morphological status,
even though they comprise the letter patterns of a stem and an affix.
2. The model of Giraudo and Grainger lacks a sublexical morphological processing
component. This extension is implemented in the model of Diependaele et al., 2005, 2009).
3. The predictors for the by-participant multiple imputations were Prime Type,
Relatedness, Trial Number, Lexicality Previous Trial, Accuracy Previous Trial, Log Prime
Frequency, Prime Length, Log Target Frequency, Target Length and Log Target Family Size.
Missing RTs were replaced because of an imbalance in the amount of non-available data
across the L1 and L2 experiments (6% versus 9% versus 11% in Experiment 1, 2 and 3,
respectively). Such an imbalance is harmful for a straightforward comparison. Since lexical
decision errors are not random (i.e., they correlate with item difficulty and participant
proficiency), simple mean imputation (and related intuitive strategies) are depreciated in favor
of regression-based multiple imputation techniques (cf. Schafer & Graham, 2002). In the
present context it turned out that the pattern of statistical results across all our experiments
remained unaltered by the imputation. For good practice, we nevertheless only report the
results from the analyses without missing values.
4. For RTs, we looked at the empirical p-value for the hypothesis that the MCMC
sample values for the 2 interaction terms in the model had a mean of zero versus a general
multivariate distribution with elliptical contours (implemented in the aovlmer.fnc function of
the languageR package; Baayen, 2009). For errors, we looked at the log-likelihood ratio
between the full and the main effects model.
!"#$%&'()*'+',-."+%/001.$#%-2%34%"25%36%
%
94%
5. We tested the model 3 times with a different reference level for Prime Type
(transparent, opaque, form) in order to obtain estimates for the individual effects. The multiple
comparison p-values were corrected following Benjamini and Hochberg (1995).
6. It might seem odd that even though proficiency is associated with an earlier age,
the age of acquisition is larger compared the Spanish-English bilinguals. There is a reasonable
explanation for this, however. In Flanders, each child is obliged to learn English at school from
the age of 12-13 onwards (up to 18 years). The average age of acquisition score of 11.97
clearly reflects this. Children are nevertheless already greatly familiarized with English via
television at this age. There are a lot of child-oriented television programs that are English
spoken with Dutch subtitles. So it is in fact quite natural that our Flemish participants report a
relatively early age of proficiency for English, even though they only received formal
education from the age of 12.
7. If missing values are ignored and condition means are analyzed in standard
repeated measures ANOVA, one implicitly replaces missing values with the condition means.
